What's the difference between git clone --mirror and git clone --bare

Question

The git clone help page only says that --mirror "Set up a mirror of the remote repository. This implies --bare." But doesn't go into detail about how the --mirror clone is different from a --bare clone.

Answer 1

$ git clone --mirror $URL

is a short-hand for

$ git clone --bare $URL
$ (cd $(basename $URL) && git remote add --mirror origin $URL)

(Copied directly from here)

How the current man-page puts it:

Compared to --bare, --mirror not only maps local branches of the source to local branches of the target, it maps all refs (including remote branches, notes etc.) and sets up a refspec configuration such that all these refs are overwritten by a git remote update in the target repository.

Answer 2

The difference is that all refs are copied as-is. This means everything: remote-tracking branches, notes, refs/originals/* (backups from filter-branch). The cloned repo has it all. It's also set up so that a remote update will re-fetch everything from the origin (overwriting the copied refs). The idea is really to mirror the repository, to have a total copy, so that you could for example host your central repo in multiple places, or back it up. Think of just straight-up copying the repo, except in a much more elegant git way.

The new documentation pretty much says all this:

--mirror

Set up a mirror of the source repository. This implies --bare. Compared to --bare, --mirror not only maps local branches of the source to local branches of the target, it maps all refs (including remote branches, notes etc.) and sets up a refspec configuration such that all these refs are overwritten by a git remote update in the target repository.

My original answer also noted the differences between a bare clone and a normal (non-bare) clone - the non-bare clone sets up remote tracking branches, only creating a local branch for HEAD, while the bare clone copies the branches directly.

Suppose origin has a few branches (master (HEAD), next, pu, and maint), some tags (v1, v2, v3), some remote branches (devA/master, devB/master), and some other refs (refs/foo/bar, refs/foo/baz, which might be notes, stashes, other devs' namespaces, who knows).

• git clone origin-url (non-bare): you will get all of the tags copied, a local branch master (HEAD) tracking a remote branch origin/master, and remote branches origin/next, origin/pu, and origin/maint. The tracking branches are set up so that if you do something like git fetch origin, they'll be fetched as you expect. Any remote branches (in the cloned remote) and other refs are completely ignored.

• git clone --bare origin-url: you will get all of the tags copied, local branches master (HEAD), next, pu, and maint, no remote tracking branches. That is, all branches are copied as is, and it's set up completely independent, with no expectation of fetching again. Any remote branches (in the cloned remote) and other refs are completely ignored.

• git clone --mirror origin-url: every last one of those refs will be copied as-is. You'll get all the tags, local branches master (HEAD), next, pu, and maint, remote branches devA/master and devB/master, other refs refs/foo/bar and refs/foo/baz. Everything is exactly as it was in the cloned remote. Remote tracking is set up so that if you run git remote update all refs will be overwritten from origin, as if you'd just deleted the mirror and recloned it. As the docs originally said, it's a mirror. It's supposed to be a functionally identical copy, interchangeable with the original.